The Open Quantum Materials Database (OQMD): assessing the accuracy of DFT formation energies

The Open Quantum Materials Database(OQMD)is a high-throughput database currently consisting of nearly 300,000 density functional theory(DFT)total energy calculations of compounds from the Inorganic Crystal Structure Database(ICSD)and decorations of commonly occurring crystal structures.To maximise the impact of these data,the entire database is being made available,without restrictions,at www.oqmd.org/download.In this paper,we outline the structure and contents of the database,and then use it to evaluate the accuracy of the calculations therein by comparing DFT predictions with experimental measurements for the stability of all elemental ground-state structures and 1,670 experimental formation energies of compounds.This represents the largest comparison between DFT and experimental formation energies to date.The apparent mean absolute error between experimental measurements and our calculations is 0.096 eV/atom.In order to estimate how much error to attribute to the DFT calculations,we also examine deviation between different experimental measurements themselves where multiple sources are available,and find a surprisingly large mean absolute error of 0.082 eV/atom.Hence,we suggest that a significant fraction of the error between DFT and experimental formation energies may be attributed to experimental uncertainties.Finally,we evaluate the stability of compounds in the OQMD(including compounds obtained from the ICSD as well as hypothetical structures),which allows us to predict the existence of~3,200 new compounds that have not been experimentally characterised and uncover trends in material discovery,based on historical data available within the ICSD.

Recent advances and applications of deep learning methods in materials science

Deep learning(DL)is one of the fastest-growing topics in materials data science,with rapidly emerging applications spanning atomistic,image-based,spectral,and textual data modalities.DL allows analysis of unstructured data and automated identification of features.The recent development of large materials databases has fueled the application of DL methods in atomistic prediction in particular.In contrast,advances in image and spectral data have largely leveraged synthetic data enabled by high-quality forward models as well as by generative unsupervised DL methods.In this article,we present a high-level overview of deep learning methods followed by a detailed discussion of recent developments of deep learning in atomistic simulation,materials imaging,spectral analysis,and natural language processing.For each modality we discuss applications involving both theoretical and experimental data,typical modeling approaches with their strengths and limitations,and relevant publicly available software and datasets.We conclude the review with a discussion of recent cross-cutting work related to uncertainty quantification in this field and a brief perspective on limitations,challenges,and potential growth areas for DL methods in materials science.

Accurate and scalable graph neural network force field and molecular dynamics with direct force architecture

Recently,machine learning(ML)has been used to address the computational cost that has been limiting ab initio molecular dynamics(AIMD).Here,we present GNNFF,a graph neural network framework to directly predict atomic forces from automatically extracted features of the local atomic environment that are translationally-invariant,but rotationally-covariant to the coordinate of the atoms.We demonstrate that GNNFF not only achieves high performance in terms of force prediction accuracy and computational speed on various materials systems,but also accurately predicts the forces of a large MD system after being trained on forces obtained from a smaller system.Finally,we use our framework to perform an MD simulation of Li7P3S11,a superionic conductor,and show that resulting Li diffusion coefficient is within 14%of that obtained directly from AIMD.The high performance exhibited by GNNFF can be easily generalized to study atomistic level dynamics of other material systems.

Scale-invariant machine-learning model accelerates the discovery of quaternary chalcogenides with ultralow lattice thermal conductivity

We design an advanced machine-learning(ML)model based on crystal graph convolutional neural network that is insensitive to volumes(i.e.,scale)of the input crystal structures to discover novel quaternary chalcogenides,AMM′Q3(A/M/M'=alkali,alkaline earth,post-transition metals,lanthanides,and Q=chalcogens).These compounds are shown to possess ultralow lattice thermal conductivity(κ_(l)),a desired requirement for thermal-barrier coatings and thermoelectrics.Upon screening the thermodynamic stability of~1 million compounds using the ML model iteratively and performing density-functional theory(DFT)calculations for a small fraction of compounds,we discover 99 compounds that are validated to be stable in DFT.Taking several DFT-stable compounds,we calculate theirκl using Peierls–Boltzmann transport equation,which reveals ultralowκ_(l)(<2 Wm^(−1)K^(−1)at room temperature)due to their soft elasticity and strong phonon anharmonicity.Our work demonstrates the high efficiency of scale-invariant ML model in predicting novel compounds and presents experimental-research opportunities with these new compounds.

Accelerated discovery of a large family of quaternary chalcogenides with very low lattice thermal conductivity

The development of efficient thermal energy management devices such as thermoelectrics and barrier coatings often relies on compounds having low lattice thermal conductivity(κl).Here,we present the computational discovery of a large family of 628 thermodynamically stable quaternary chalcogenides,AMM′Q_(3)(A=alkali/alkaline earth/post-transition metals;M/M′=transition metals,lanthanides;Q=chalcogens)using high-throughput density functional theory(DFT)calculations.We validate the presence of lowκl in these materials by calculatingκl of several predicted stable compounds using the Peierls–Boltzmann transport equation.Our analysis reveals that the lowκl originates from the presence of either a strong lattice anharmonicity that enhances the phononscatterings or rattler cations that lead to multiple scattering channels in their crystal structures.Our thermoelectric calculations indicate that some of the predicted semiconductors may possess high energy conversion efficiency with their figure-of-merits exceeding 1 near 600 K.Our predictions suggest experimental research opportunities in the synthesis and characterization of these stable,low κ_(l) compounds.

Microscopic mechanism of unusual lattice thermal transport in TlInTe_(2)

We investigate the microscopic mechanism of ultralow lattice thermal conductivity(κl)of TlInTe_(2)and its weak temperature dependence using a unified theory of lattice heat transport,that considers contributions arising from the particle-like propagation as well as wave-like tunneling of phonons.While we use the Peierls–Boltzmann transport equation(PBTE)to calculate the particlelike contributions(κl(PBTE)),we explicitly calculate the off-diagonal(OD)components of the heat-flux operator within a firstprinciples density functional theory framework to determine the contributions(κl(OD))arising from the wave-like tunneling of phonons.At each temperature,T,we anharmonically renormalize the phonon frequencies using the self-consistent phonon theory including quartic anharmonicity,and utilize them to calculateκl(PBTE)andκl(OD).With the combined inclusion ofκl(PBTE),κl(OD),and additional grain-boundary scatterings,our calculations successfully reproduce the experimental results.Our analysis shows that large quartic anharmonicity of TlInTe_(2)(a)strongly hardens the low-energy phonon branches,(b)diminishes the three-phonon scattering processes at finite T,and(c)recovers the weaker than T^(−1) decay of the measuredκl.

Identification of high-dielectric constant compounds from statistical design

The discovery of high-dielectric materials is crucial to increasing the efficiency of electronic devices and batteries.Here,we report three previously unexplored materials with very high dielectric constants(69<ϵ<101)and large band gaps(2.9<E_(g)(eV)<5.5)obtained by screening materials databases using statistical optimization algorithms aided by artificial neural networks(ANN).Two of these new dielectrics are mixed-anion compounds(Eu_(5)SiCl_(6)O_(4)and HoClO)and are shown to be thermodynamically stable against common semiconductors via phase diagram analysis.We also uncovered four other materials with relatively large dielectric constants(20<ϵ<40)and band gaps(2.3<E_(g)(eV)<2.7).While the ANN training-data are obtained from the Materials Project,the search-space consists of materials from the Open Quantum Materials Database(OQMD)—demonstrating a successful implementation of cross-database materials design.Overall,we report the dielectric properties of 17 materials calculated using ab initio calculations,that were selected in our design workflow.The dielectric materials with high-dielectric properties predicted in this work open up further experimental research opportunities.

A Convergent Understanding of Charged Defects

CONSPECTUS:Historically,defects in semiconductors and ionic conductors have been studied using very different approaches.In the solid-state ionics community,nonstoichiometry and defect thermochemistry are often probed directly through experiments.The dependency of defect concentrations on chemical conditions(typically oxygen pressure)are modeled using a physical chemistry framework and compactly represented by the well-known Brouwer diagrams.